Augmented Conditioning Is Enough For Effective Training Image Generation

Image generation abilities of text-to-image diffusion models have significantly advanced, yielding highly photo-realistic images from descriptive text and increasing the viability of leveraging synthetic images to train computer vision models. To serve as effective training data, generated images must be highly realistic while also sufficiently diverse within the support of the target data distribution. Yet, state-of-the-art conditional image generation models have been primarily optimized for creative applications, prioritizing image realism and prompt adherence over conditional diversity. In this paper, we investigate how to improve the diversity of generated images with the goal of increasing their effectiveness to train downstream image classification models, without fine-tuning the image generation model. We find that conditioning the generation process on an augmented real image and text prompt produces generations that serve as effective synthetic datasets for downstream training. Conditioning on real training images contextualizes the generation process to produce images that are in-domain with the real image distribution, while data augmentations introduce visual diversity that improves the performance of the downstream classifier. We validate augmentation-conditioning on a total of five established long-tail and few-shot image classification benchmarks and show that leveraging augmentations to condition the generation process results in consistent improvements over the state-of-the-art on the long-tailed benchmark and remarkable gains in extreme few-shot regimes of the remaining four benchmarks. These results constitute an important step towards effectively leveraging synthetic data for downstream training.

Can Text-to-Image Generative Models Accurately Depict Age? A Comparative Study on Synthetic Portrait Generation and Age Estimation

Text-to-image generative models have shown remarkable progress in producing diverse and photorealistic outputs. In this paper, we present a comprehensive analysis of their effectiveness in creating synthetic portraits that accurately represent various demographic attributes, with a special focus on age, nationality, and gender. Our evaluation employs prompts specifying detailed profiles (e.g., Photorealistic selfie photo of a 32-year-old Canadian male), covering a broad spectrum of 212 nationalities, 30 distinct ages from 10 to 78, and balanced gender representation. We compare the generated images against ground truth age estimates from two established age estimation models to assess how faithfully age is depicted. Our findings reveal that although text-to-image models can consistently generate faces reflecting different identities, the accuracy with which they capture specific ages and do so across diverse demographic backgrounds remains highly variable. These results suggest that current synthetic data may be insufficiently reliable for high-stakes age-related tasks requiring robust precision, unless practitioners are prepared to invest in significant filtering and curation. Nevertheless, they may still be useful in less sensitive or exploratory applications, where absolute age precision is not critical.

High-Fidelity Human Avatars from Laptop Webcams using Edge Compute

Applications of generating photo-realistic human avatars are many, however, high-fidelity avatar generation traditionally required expensive professional camera rigs and artistic labor, but recent research has enabled constructing them automatically from smartphones with RGB and IR sensors. However, these new methods still rely on the presence of high-resolution cameras on modern smartphones and often require offloading the processing to powerful servers with GPUs. Modern applications such as video conferencing call for the ability to generate these avatars from consumer-grade laptop webcams using limited compute available on-device. In this work, we develop a novel method based on 3D morphable models, landmark detection, photo-realistic texture GANs, and differentiable rendering to tackle the problem of low webcam image quality and edge computation. We build an automatic system to generate high-fidelity animatable avatars under these limitations, leveraging the neural compute capabilities of mobile chips.

Towards Consistent and Controllable Image Synthesis for Face Editing

Current face editing methods mainly rely on GAN-based techniques, but recent focus has shifted to diffusion-based models due to their success in image reconstruction. However, diffusion models still face challenges in manipulating fine-grained attributes and preserving consistency of attributes that should remain unchanged. To address these issues and facilitate more convenient editing of face images, we propose a novel approach that leverages the power of Stable-Diffusion models and crude 3D face models to control the lighting, facial expression and head pose of a portrait photo. We observe that this task essentially involve combinations of target background, identity and different face attributes. We aim to sufficiently disentangle the control of these factors to enable high-quality of face editing. Specifically, our method, coined as RigFace, contains: 1) A Spatial Arrtibute Encoder that provides presise and decoupled conditions of background, pose, expression and lighting; 2) An Identity Encoder that transfers identity features to the denoising UNet of a pre-trained Stable-Diffusion model; 3) An Attribute Rigger that injects those conditions into the denoising UNet. Our model achieves comparable or even superior performance in both identity preservation and photorealism compared to existing face editing models.

End-to-end Training for Text-to-Image Synthesis using Dual-Text Embeddings

Text-to-Image (T2I) synthesis is a challenging task that requires modeling complex interactions between two modalities ( i.e., text and image). A common framework adopted in recent state-of-the-art approaches to achieving such multimodal interactions is to bootstrap the learning process with pre-trained image-aligned text embeddings trained using contrastive loss. Furthermore, these embeddings are typically trained generically and reused across various synthesis models. In contrast, we explore an approach to learning text embeddings specifically tailored to the T2I synthesis network, trained in an end-to-end fashion. Further, we combine generative and contrastive training and use two embeddings, one optimized to enhance the photo-realism of the generated images, and the other seeking to capture text-to-image alignment. A comprehensive set of experiments on three text-to-image benchmark datasets (Oxford-102, Caltech-UCSD, and MS-COCO) reveal that having two separate embeddings gives better results than using a shared one and that such an approach performs favourably in comparison with methods that use text representations from a pre-trained text encoder trained using a discriminative approach. Finally, we demonstrate that such learned embeddings can be used in other contexts as well, such as text-to-image manipulation.

Generating customized prompts for Zero-Shot Rare Event Medical Image Classification using LLM

Rare events, due to their infrequent occurrences, do not have much data, and hence deep learning techniques fail in estimating the distribution for such data. Open-vocabulary models represent an innovative approach to image classification. Unlike traditional models, these models classify images into any set of categories specified with natural language prompts during inference. These prompts usually comprise manually crafted templates (e.g., 'a photo of a {}') that are filled in with the names of each category. This paper introduces a simple yet effective method for generating highly accurate and contextually descriptive prompts containing discriminative characteristics. Rare event detection, especially in medicine, is more challenging due to low inter-class and high intra-class variability. To address these, we propose a novel approach that uses domain-specific expert knowledge on rare events to generate customized and contextually relevant prompts, which are then used by large language models for image classification. Our zero-shot, privacy-preserving method enhances rare event classification without additional training, outperforming state-of-the-art techniques.

CE-SDWV: Effective and Efficient Concept Erasure for Text-to-Image Diffusion Models via a Semantic-Driven Word Vocabulary

Large-scale text-to-image (T2I) diffusion models have achieved remarkable generative performance about various concepts. With the limitation of privacy and safety in practice, the generative capability concerning NSFW (Not Safe For Work) concepts is undesirable, e.g., producing sexually explicit photos, and licensed images. The concept erasure task for T2I diffusion models has attracted considerable attention and requires an effective and efficient method. To achieve this goal, we propose a CE-SDWV framework, which removes the target concepts (e.g., NSFW concepts) of T2I diffusion models in the text semantic space by only adjusting the text condition tokens and does not need to re-train the original T2I diffusion model's weights. Specifically, our framework first builds a target concept-related word vocabulary to enhance the representation of the target concepts within the text semantic space, and then utilizes an adaptive semantic component suppression strategy to ablate the target concept-related semantic information in the text condition tokens. To further adapt the above text condition tokens to the original image semantic space, we propose an end-to-end gradient-orthogonal token optimization strategy. Extensive experiments on I2P and UnlearnCanvas benchmarks demonstrate the effectiveness and efficiency of our method.

Continuous 3D Perception Model with Persistent State

We present a unified framework capable of solving a broad range of 3D tasks. Our approach features a stateful recurrent model that continuously updates its state representation with each new observation. Given a stream of images, this evolving state can be used to generate metric-scale pointmaps (per-pixel 3D points) for each new input in an online fashion. These pointmaps reside within a common coordinate system, and can be accumulated into a coherent, dense scene reconstruction that updates as new images arrive. Our model, called CUT3R (Continuous Updating Transformer for 3D Reconstruction), captures rich priors of real-world scenes: not only can it predict accurate pointmaps from image observations, but it can also infer unseen regions of the scene by probing at virtual, unobserved views. Our method is simple yet highly flexible, naturally accepting varying lengths of images that may be either video streams or unordered photo collections, containing both static and dynamic content. We evaluate our method on various 3D/4D tasks and demonstrate competitive or state-of-the-art performance in each. Project Page: https://cut3r.github.io/

BlanketGen2-Fit3D: Synthetic Blanket Augmentation Towards Improving Real-World In-Bed Blanket Occluded Human Pose Estimation

Human Pose Estimation (HPE) from monocular RGB images is crucial for clinical in-bed skeleton-based action recognition, however, it poses unique challenges for HPE models due to the frequent presence of blankets occluding the person, while labeled HPE data in this scenario is scarce. To address this we introduce BlanketGen2-Fit3D (BG2-Fit3D), an augmentation of Fit3D dataset that contains 1,217,312 frames with synthetic photo-realistic blankets. To generate it we used BlanketGen2, our new and improved version of our BlanketGen pipeline that simulates synthetic blankets using ground-truth Skinned Multi-Person Linear model (SMPL) meshes and then renders them as transparent images that can be layered on top of the original frames. This dataset was used in combination with the original Fit3D to finetune the ViTPose-B HPE model, to evaluate synthetic blanket augmentation effectiveness. The trained models were further evaluated on a real-world blanket occluded in-bed HPE dataset (SLP dataset). Comparing architectures trained on only Fit3D with the ones trained with our synthetic blanket augmentation the later improved pose estimation performance on BG2-Fit3D, the synthetic blanket occluded dataset significantly to (0.977 Percentage of Correct Keypoints (PCK), 0.149 Normalized Mean Error (NME)) with an absolute 4.4% PCK increase. Furthermore, the test results on SLP demonstrated the utility of synthetic data augmentation by improving performance by an absolute 2.3% PCK, on real-world images with the poses occluded by real blankets. These results show synthetic blanket augmentation has the potential to improve in-bed blanket occluded HPE from RGB images. The dataset as well as the code will be made available to the public.

GPS as a Control Signal for Image Generation

We show that the GPS tags contained in photo metadata provide a useful control signal for image generation. We train GPS-to-image models and use them for tasks that require a fine-grained understanding of how images vary within a city. In particular, we train a diffusion model to generate images conditioned on both GPS and text. The learned model generates images that capture the distinctive appearance of different neighborhoods, parks, and landmarks. We also extract 3D models from 2D GPS-to-image models through score distillation sampling, using GPS conditioning to constrain the appearance of the reconstruction from each viewpoint. Our evaluations suggest that our GPS-conditioned models successfully learn to generate images that vary based on location, and that GPS conditioning improves estimated 3D structure.